ThoughtForge AI

Our EIR's recently attended Automate 2024 to stay up-to-date on the latest trends in automation. Check out our latest blog post for a recap of the event and to learn more about the future of automation. PROPHESEE Micropsi Industries UnitX MūL Technologies Multiway Robotics SNOWBOTIX ThoughtForge AI Pittsburgh Robotics Network Interactive Training Systems, Inc. Mech-Mind Robotics Mobile Industrial Robots Exotec ForwardX Robotics https://hubs.la/Q02xfYtx0 #Automate2024 #automation #innovation

It was so fun catching up with Matthew Brown and Harneet S. at the ThoughtForge AI booth last week at #automate2024. I recorded this impressive demo there. Have you ever heard of Active Inference? It's an #ai technique that allows #artificialintelligence (in this case controlling a Kuka robot) to learn tasks extremely quickly and transfer expertise from one skill to another similar skill. Let me explain what you are seeing here. The board on the left contains 16 different shaped bolts, knobs, and screws. The AI trained on one of those bolts only. And it only made three total attempts to turn the knob (that's not much practice)! I picked a different knob (the most differently shaped one I could find) and what you are seeing is Harneet moving the whole board around to make the task even more difficult. The agent skill found and turned the knob I selected, correctly, the first time. #transferlearning #activeinference #intelligentagents #robotics

There’s a lot of AI hype these days. There is a lot of really great technology coming out everyday. It’s hard to understand what to use, when and why if you’re not AI nerds like us. Active Inference (AIF) is still a relatively new field in neuroscience and there are only three companies using the approach for modeling AI (that we know of) and one (us) with publicly available models to test. And we made a number of innovations to traditional approaches (NetAIF) which enable us to use less data to train our models, making them very small (less than 1 CPU) and able to continue learning post-deployment in real-time (1-2 milliseconds) for on going accuracy. A lot of people ask us how we’re different than LLMs, LAMs, Diffusion Models and other approaches to AI. For the non-engineers and non-scientists in the audience, the way we like to explain the difference is like this: Imagine solving a Rubix cube: - LLMs are great for identifying the instructional path to solve the cube - LAMs would translate those instructions into physical actions - Diffusion models would imitate what a human would do - ThoughtForge’s NetAIF would be what you’d use when the cube gets stuck and you need to jiggle it to turn - for those unpredictable scenarios when you need real time adaptive dexterous control and responsiveness If you’re curious and want to learn more shoot any of us a note, we love talking about it and are happy to share.

At Automate in Chicago? Stop by our booth to see two interactive demos. Robotic AI with no WiFi!! Booth #4490

Come check out ThoughtForge AI Automate Show at Booth 4490. Matthew Brown, Harneet S. and I will be demoing two of ThoughtForge’s adaptive AI models. We’d love to show you what we’ve been up to.

Come visit ThoughtForge’s booth at #hannovermesse : We are doing live interactive demos and would love to share more about our real-time adaptive robot AI!

"Initially, you might expect that perception occurs in the following steps: your sensory organs, like your eyes and ears, start by taking in some information about the world. The brain then uses that information to create your internal, cohesive image of the world — your conscious experience. However... your perception of the world occurs more like this: the brain continuously generates internal models, or best guesses, about what the external world should look like at any given moment based on past experiences and stored knowledge. It then compares these predictions with incoming sensory data. When there’s a mismatch between prediction and actual sensory input — a prediction error — the brain updates its models to reduce this error. In this way, what you perceive at any moment is not a direct representation of the outside world but rather the brain’s interpretation or hypothesis of that world, constantly refined by prediction errors. " https://lnkd.in/g72zJU4G

"Initially, you might expect that perception occurs in the following steps: your sensory organs, like your eyes and ears, start by taking in some information about the world. The brain then uses that information to create your internal, cohesive image of the world — your conscious experience. However... your perception of the world occurs more like this: the brain continuously generates internal models, or best guesses, about what the external world should look like at any given moment based on past experiences and stored knowledge. It then compares these predictions with incoming sensory data. When there’s a mismatch between prediction and actual sensory input — a prediction error — the brain updates its models to reduce this error. In this way, what you perceive at any moment is not a direct representation of the outside world but rather the brain’s interpretation or hypothesis of that world, constantly refined by prediction errors. " https://lnkd.in/g72zJU4G

"Chronic worker shortages have led many companies to invest in machines to do some of the work they can’t find people to do. They’ve also been training the workers they do have to use advanced technology so they can produce more with less. The result has been an unexpected productivity boom, which helps explain a great economic mystery: How has the world’s largest economy managed to remain so healthy, with brisk growth and low unemployment, despite brutally high interest rates that are intended to tame inflation but that typically cause a recession?" https://lnkd.in/eGM6JP7x